Advancing Set-Conditional Set Generation: Graph Diffusion for Fast Simulation of Reconstructed Particles

The computational intensity of detailed detector simulations poses a significant bottleneck in generating simulated data for collider experiments. This challenge inspires the continued development of fast simulation techniques based on machine learning to serve as efficient surrogate models. In our approach, a network generates a set of reconstructed objects conditioned on input particle sets. Building on the success of a slot-attention-based model, we present a new architecture utilizing diffusion, showcasing an enhanced performance in the context of single jets.